船用离心泵故障SOM网络诊断方法

Fault Diagnosis of Marine Centrifugal Pump with the SOM Network

为了实现对船用离心泵的实时在线智能故障诊断,进行了基于SOM网络(自组织特征映射神经网络)的船用离心泵故障诊断方法研究。在分析船用离心泵典型故障及特征的基础上,建立故障模型,提取故障特征向量并建立学习样本;设计和组建了SOM神经网络,将SOM网络的抽取输入信号模式特征的能力应用于故障诊断;通过网络训练建立了SOM网络输入与输出属性间良好的非线性映射,实现了将特征向量输入网络来诊断故障。经实验验证,该方法具有良好的准确度和适应性。

The fault diagnosis technology for the marine centrifugal pump based on the SOM network （self-organi- zing feature map neural network） is studied. On the basis of analyzing typical failures and corresponding characteris- tics of marine centrifugal pumps, a failure model is built up; the fault feature vectors are extracted; and the training samples are established. The SOM neural network, suitable for fault diagnosis due to its signal feature extraction a- bility, is designed and constructed. A non linear mapping between SOM network input and output vectors for the fault diagnose is established by means of network training. The diagnosis is output when marine centrifugal pump feature vectors are input to the networks. According to verification tests, this method has good accuracy and adaptability.